Systems and methods for recommending cold-start items on a website of a retailer

Abstract

Systems and methods including one or more processing modules and one or more non-transitory storage modules storing computing instructions configured to run on the one or more processing modules and perform acts of training one or more first models to recommend a first item after a user has had an interaction on the web site of the online retailer with a second item, determining static features common to both the first item and the second item, training a second model to determine whether to coordinate a display of any new item as one of one or more recommended items with any of a plurality of items, and coordinating the display of the new item as one of the one or more recommended items when the one or more of the plurality of items are displayed on the website of the online retailer based on the static features of the new item.

What is claimed is: 
     
       1. A system comprising:
 one or more processing modules; and 
 one or more non-transitory storage modules storing computing instructions configured to run on the one or more processing modules and perform acts of:
 training, using a historical traffic pattern of a plurality of items on a website of an online retailer, one or more first models to recommend one or more first items of the plurality of items for sale on the website of the online retailer after a user has had an interaction on the website of the online retailer with one or more second items of the plurality of items; 
 determining static features common to both the one or more first items and the one or more second items; 
 training, using the static features of the one or more first items and the one or more second items, a second model to determine whether to coordinate a display of any new item as one of one or more recommended items with any of the plurality of items; 
 determining, using the second model and one or more static features of the new item, whether to coordinate a display of the new item as one of the one or more recommended items when one or more of the plurality of items are displayed on the website of the online retailer; and 
 coordinating the display of the new item as the one of the one or more recommended items when the one or more of the plurality of items are displayed on the website of the online retailer based on using the second model and the one or more static features of the new item. 
 
 
     
     
       2. The system of  claim 1 , wherein determining the static features comprises determining, using ensemble learning, the static features common to both the one or more first items and the one or more second items, the static features comprising one or more of a title, a description, an image, a price, a category, or item specifications. 
     
     
       3. The system of  claim 2 , wherein the ensemble learning comprises a blend of a set of models, the set of models comprising a linear model, a matrix factorization model, and a neural network model. 
     
     
       4. The system of  claim 1 , wherein the one or more non-transitory storage modules storing the computing instructions are further configured to run on the one or more processing modules and perform acts of:
 labelling as positive training labels one or more existing recommendations of the one or more first items after the user has had the interaction on the website of the online retailer with the one or more second items; and 
 labelling as negative training labels additional items of the plurality of items not recommended after the user has had the interaction on the website of the online retailer with the one or more second items. 
 
     
     
       5. The system of  claim 1 , wherein the interaction with the one or more second items comprises at least one of:
 a viewed-ultimately-bought (VUB) interaction wherein the user had viewed the one or more second items and ultimately bought the one or more first items; 
 a viewed-also-viewed (VAV) interaction wherein the user had viewed the one or more second items and also viewed the one or more first items during a single browsing session; and 
 a bought-also-bought (BAB) interaction wherein the user had bought the one or more second items and also bought the one or more first items within a predetermined window of time. 
 
     
     
       6. The system of  claim 5 , wherein the one or more first models comprise at least three first models comprising:
 a VUB model configured to make a first recommendation of the one or more first items based on the VUB interaction of the user with the one or more second items and the one or more first items, wherein the first recommendation is based on a combination of (1) the one or more first items substituting the one or more second items, and (2) the one or more first items complementing the one or more second items; 
 a VAV model configured to make a second recommendation of the one or more first items based on the VAV interaction of the user with the one or more second items and the one or more first items, wherein the second recommendation is based on the one or more first items substituting the one or more second items; and 
 a BAB model configured to make a third recommendation of the one or more first items based on the BAB interaction of the user with the one or more second items and the one or more first items, wherein the third recommendation is based on the one or more first items complementing the one or more second items. 
 
     
     
       7. The system of  claim 1 , wherein training the second model comprises training a probabilistic model to combine scoring from the one or more first models and determine whether to coordinate the display of the any new item as the one of the one or more recommended items with the any of the plurality of items. 
     
     
       8. The system of  claim 1 , wherein the one or more non-transitory storage modules storing the computing instructions are further configured to run on the one or more processing modules and perform acts of:
 determining a ranking of a plurality of new items based on a likelihood of an additional user selecting each new item of the plurality of new items when each new item is displayed as the one of the one or more recommended items when one or more of the plurality of items are displayed on the website of the online retailer; 
 determining a quality of a seller of each new item of the plurality of new items; and 
 adjusting the ranking of the plurality of new items based on the quality of the seller of each new item of the plurality of new items. 
 
     
     
       9. The system of  claim 1 , wherein the one or more non-transitory storage modules storing the computing instructions are further configured to run on the one or more processing modules and perform acts of:
 collecting feedback comprising statistics of how many users selected the display of the new item when the new item was displayed as the one of the one or more recommended items when the one or more of the plurality of items were displayed on the website; and 
 adjusting the second model through reinforcement learning based on the feedback. 
 
     
     
       10. The system of  claim 1 , wherein:
 determining the static features comprises determining, using ensemble learning, the static features common to both the one or more first items and the one or more second items, the static features comprising one or more of a title, a description, an image, a price, a category, or item specifications, and the ensemble learning comprising a blend of a set of models, the set of models comprising a linear model, a matrix factorization model, and a neural network model; 
 the one or more non-transitory storage modules storing the computing instructions are further configured to run on the one or more processing modules and perform acts of:
 labelling as positive training labels one or more existing recommendations of the one or more first items after the user has had the interaction on the website of the online retailer with the one or more second items; and 
 labelling as negative training labels additional items of the plurality of items not recommended after the user has had the interaction on the website of the online retailer with the one or more second items; 
 
 the interaction with the one or more second items comprises at least one of:
 a viewed-ultimately-bought (VUB) interaction wherein the user had viewed the one or more second items and ultimately bought the one or more first items; 
 a viewed-also-viewed (VAV) interaction wherein the user had viewed the one or more second items and also viewed the one or more first items during a single browsing session; and 
 a bought-also-bought (BAB) interaction wherein the user had bought the one or more second items and also bought the one or more first items within a predetermined window of time; 
 
 the one or more first models comprise at least three first models comprising:
 a VUB model configured to make a first recommendation of the one or more first items based on the VUB interaction of the user with the one or more second items and the one or more first items, wherein the first recommendation is based on a combination of (1) the one or more first items substituting the one or more second items, and (2) the one or more first items complementing the one or more second items; 
 a VAV model configured to make a second recommendation of the one or more first items based on the VAV interaction of the user with the one or more second items and the one or more first items, wherein the second recommendation is based on the one or more first items substituting the one or more second items; and 
 a BAB model configured to make a third recommendation of the one or more first items based on the BAB interaction of the user with the one or more second items and the one or more first items, wherein the third recommendation is based on the one or more first items complementing the one or more second items; 
 
 training the second model comprises training a probabilistic model to combine scoring from the one or more first models and determine whether to coordinate the display of the any new item as the one of the one or more recommended items with the any of the plurality of items; and 
 the one or more non-transitory storage modules storing the computing instructions are further configured to run on the one or more processing modules and perform acts of:
 determining a ranking of a plurality of new items based on a likelihood of an additional user selecting each new item of the plurality of new items when each new item is displayed as the one of the one or more recommended items when one or more of the plurality of items are displayed on the website of the online retailer; 
 determining a quality of a seller of each new item of the plurality of new items; 
 adjusting the ranking of the plurality of new items based on the quality of the seller of each new item of the plurality of new items; 
 collecting feedback comprising statistics of how many users selected the display of the new item when the new item was displayed as the one of the one or more recommended items when the one or more of the plurality of items were displayed on the website; and 
 adjusting the second model through reinforcement learning based on the feedback. 
 
 
     
     
       11. A method comprising:
 training, using a historical traffic pattern of a plurality of items on a website of an online retailer, one or more first models to recommend one or more first items of the plurality of items for sale on the website of the online retailer after a user has had an interaction on the website of the online retailer with one or more second items of the plurality of items; 
 determining static features common to both the one or more first items and the one or more second items; 
 training, using the static features of the one or more first items and the one or more second items, a second model to determine whether to coordinate a display of any new item as one of one or more recommended items with any of the plurality of items; 
 determining, using the second model and one or more static features of the new item, whether to coordinate a display of the new item as one of the one or more recommended items when one or more of the plurality of items are displayed on the website of the online retailer; and 
 coordinating the display of the new item as the one of the one or more recommended items when the one or more of the plurality of items are displayed on the website of the online retailer based on using the second model and the one or more static features of the new item. 
 
     
     
       12. The method of  claim 11 , wherein determining the static features comprises determining, using ensemble learning, the static features common to both the one or more first items and the one or more second items, the static features comprising one or more of a title, a description, an image, a price, a category, or item specifications. 
     
     
       13. The method of  claim 12 , wherein the ensemble learning comprises a blend of a set of models, the set of models comprising a linear model, a matrix factorization model, and a neural network model. 
     
     
       14. The method of  claim 11 , further comprising:
 labelling as positive training labels one or more existing recommendations of the one or more first items after the user has had the interaction on the website of the online retailer with the one or more second items; and 
 labelling as negative training labels additional items of the plurality of items not recommended after the user has had the interaction on the website of the online retailer with the one or more second items. 
 
     
     
       15. The method of  claim 11 , wherein the interaction with the one or more second items comprises at least one of:
 a viewed-ultimately-bought (VUB) interaction wherein the user had viewed the one or more second items and ultimately bought the one or more first items; 
 a viewed-also-viewed (VAV) interaction wherein the user had viewed the one or more second items and also viewed the one or more first items during a single browsing session; and 
 a bought-also-bought (BAB) interaction wherein the user had bought the one or more second items and also bought the one or more first items within a predetermined window of time. 
 
     
     
       16. The method of  claim 15 , wherein the one or more first models comprise at least three first models comprising:
 a VUB model configured to make a first recommendation of the one or more first items based on the VUB interaction of the user with the one or more second items and the one or more first items, wherein the first recommendation is based on a combination of (1) the one or more first items substituting the one or more second items, and (2) the one or more first items complementing the one or more second items; 
 a VAV model configured to make a second recommendation of the one or more first items based on the VAV interaction of the user with the one or more second items and the one or more first items, wherein the second recommendation is based on the one or more first items substituting the one or more second items; and 
 a BAB model configured to make a third recommendation of the one or more first items based on the BAB interaction of the user with the one or more second items and the one or more first items, wherein the third recommendation is based on the one or more first items complementing the one or more second items. 
 
     
     
       17. The method of  claim 11 , wherein training the second model comprises training a probabilistic model to combine scoring from the one or more first models and determine whether to coordinate the display of the any new item as the one of the one or more recommended items with the any of the plurality of items. 
     
     
       18. The method of  claim 11 , further comprising:
 determining a ranking of a plurality of new items based on a likelihood of an additional user selecting each new item of the plurality of new items when each new item is displayed as the one of the one or more recommended items when one or more of the plurality of items are displayed on the website of the online retailer; 
 determining a quality of a seller of each new item of the plurality of new items; and 
 adjusting the ranking of the plurality of new items based on the quality of the seller of each new item of the plurality of new items. 
 
     
     
       19. The method of  claim 11 , further comprising:
 collecting feedback comprising statistics of how many users selected the display of the new item when the new item was displayed as the one of the one or more recommended items when the one or more of the plurality of items were displayed on the web site; and 
 adjusting the second model through reinforcement learning based on the feedback. 
 
     
     
       20. The method of  claim 11 , wherein:
 determining the static features comprises determining, using ensemble learning, the static features common to both the one or more first items and the one or more second items, the static features comprising one or more of a title, a description, an image, a price, a category, or item specifications, and the ensemble learning comprising a blend of a set of models, the set of models comprising a linear model, a matrix factorization model, and a neural network model; 
 the method further comprises:
 labelling as positive training labels one or more existing recommendations of the one or more first items after the user has had the interaction on the website of the online retailer with the one or more second items; and 
 labelling as negative training labels additional items of the plurality of items not recommended after the user has had the interaction on the website of the online retailer with the one or more second items; 
 
 the interaction with the one or more second items comprises at least one of:
 a viewed-ultimately-bought (VUB) interaction wherein the user had viewed the one or more second items and ultimately bought the one or more first items; 
 a viewed-also-viewed (VAV) interaction wherein the user had viewed the one or more second items and also viewed the one or more first items during a single browsing session; and 
 a bought-also-bought (BAB) interaction wherein the user had bought the one or more second items and also bought the one or more first items within a predetermined window of time; 
 
 the one or more first models comprise at least three first models comprising:
 a VUB model configured to make a first recommendation of the one or more first items based on the VUB interaction of the user with the one or more second items and the one or more first items, wherein the first recommendation is based on a combination of (1) the one or more first items substituting the one or more second items, and (2) the one or more first items complementing the one or more second items; 
 a VAV model configured to make a second recommendation of the one or more first items based on the VAV interaction of the user with the one or more second items and the one or more first items, wherein the second recommendation is based on the one or more first items substituting the one or more second items; and 
 a BAB model configured to make a third recommendation of the one or more first items based on the BAB interaction of the user with the one or more second items and the one or more first items, wherein the third recommendation is based on the one or more first items complementing the one or more second items; 
 
 training the second model comprises training a probabilistic model to combine scoring from the one or more first models and determine whether to coordinate the display of the any new item as the one of the one or more recommended items with the any of the plurality of items; and 
 the method further comprises:
 determining a ranking of a plurality of new items based on a likelihood of an additional user selecting each new item of the plurality of new items when each new item is displayed as the one of the one or more recommended items when one or more of the plurality of items are displayed on the website of the online retailer; 
 determining a quality of a seller of each new item of the plurality of new items; 
 adjusting the ranking of the plurality of new items based on the quality of the seller of each new item of the plurality of new items; 
 collecting feedback comprising statistics of how many users selected the display of the new item when the new item was displayed as the one of the one or more recommended items when the one or more of the plurality of items were displayed on the website; and 
 adjusting the second model through reinforcement learning based on the feedback.